The detection method described in this paper uses autoencoder-based feature learning to identify abnormal traffic patterns indicative of a Mozi infection and employs hashing techniques to track and enumerate the P2P botnet nodes. The Internet of Things (IoT) has emerged as a game-changer in today’s world, influencing numerous industries and lifelines. Detection of IoT botnets has multiple implications for the security and availability of IoT ecosystems. Mozi is a P2P IoT botnet with characteristics such as fast spreading, persistence, and misuse of weak device configurations. Traditional signature and rule-based detection schemes are often unable to detect such dynamic threats, due to their poor generalization capabilities. In this paper, we present an efficient Mozi botnet detection scheme that leverages auto encoder-based feature learning and a hashing technique to achieve fast, scalable detection. Therefore, this paper proposes a method to detect botnets using an autoencoder and a hash function for large-scale data retrieval. The algorithm uses the autoencoder to learn what is normal, then it hashes what it learned to succinctly summarize the learned information, and it compares hash codes in real time to detect anomalies, including IoT botnet-related anomalies. The proposed method achieves high detection accuracy, and is robust to the evolving attack strategies of Mozi botnet, better than traditional methods with the low computation and storage overhead, which can be well applied in IoT infrastructure.

Graphics Processing Units (GPUs) used in safety-critical applications are growing, but their massively parallel architectures introduce vulnerabilities that allow side-channel attacks to steal secret information. Previous studies have established that the timing, contention, power, and access pattern side-channel are feasible against GPU workloads. However, their existing defences are still inadequate to protect real-world shader executions systematically. A new and inventive game-theoretical methodology is suggested to simulate and appraise the complex interplay between attackers and defenders in GPU side-channel attacks. The interactions are recast as a two-player, non-cooperative game, and the optimal strategies of both players have been determined under different payoff models and threat scenarios. Results from experiments conducted on commercial GPUs further validate that the proposed method accurately depicts adversarial dynamics in real life, urging the creation of robust countermeasures. This research aims to close the gap between theoretical security analysis and pragmatic GPU defence mechanisms. We give a rigid base to design suitable and safe GPU architectures for constantly evolving side channel threadings. Driving work through humans follows a human-centred point of view, insisting that thinking about human factors like perception, judgement, and decision-making is essential to analyse adversarial strategies in the cybersecurity domain. The game theoretical model provides a systematic framework for predicting the most likely attack vectors, evaluating defence strategies, and developing robust countermeasures tailored to the adversarial environment.